Nice to virtually meet you today. Just a quick note to let you know:
- I just posted 1) updated slides that correct the typo that I had in the tree; 2) a recording of today's lecture; 3) answers to the standardization exercise that we did in class.
- I'll post an additional, optional standardization exercise by end of day tomorrow.
In addition, one clarification: it isn't generally possible to wholly validate the exchangeability assumption based on observed data outside the context of an ideal randomized experiment. We rather use subject matter knowledge to determine whether marginal or conditional exchangeability is likely to hold. Are there variables that likely confound the relationship between the exposure and outcome? Have the investigators accounted for them? In practice, investigators also often apply different statistical approaches/modeling to assess the degree of potential confounding bias (or lack of exchangeability). There often isn't a correct answer to the question of whether or not exchangeability is likely to hold, so much as an imperative to consider this potential for bias and defend your position.
Let me know if you have any questions.
Best,
Rebecca
